阅读说明：本技术 识别带口音的语音 (Recognizing accented speech ) 是由 K·A·格雷 于 2014-01-24 设计创作，主要内容包括：本发明涉及识别带口音的语音。描述了用于识别带口音的语音的技术(300,400,500)和装置(100,200,700)。在一些实施例中,口音模块使用基于设备数据的口音库来识别(308)带口音的语音、基于识别的词语被设置为要提供到其中的应用字段而使用不同的语音识别校正水平、或者基于对未正确识别的语音做出的校正来更新(310)口音库。(The invention relates to recognizing accented speech. Techniques (300, 400, 500) and apparatus (100, 200, 700) for recognizing accented speech are described. In some embodiments, the accent module identifies (308) accented speech using an accent library based on the device data, updates (310) the accent library using different levels of speech recognition correction based on application fields into which the identified terms are set to be provided, or based on corrections made to incorrectly identified speech.)

1. A method, comprising:

receiving, at data processing hardware, audio data of an utterance captured by a computing device;

identifying, by the data processing hardware, a native language of a speaker of the utterance based on a pitch or intonation within audio data of the utterance;

selecting, by the data processing hardware, an accent library comprising phonemes of pronunciations of words for a particular language based on a native language of a speaker of the recognized utterance; and

generating, by the data processing hardware, a transcription of the utterance using a speech recognition engine modified by the selected accent library.

2. The method of claim 1, further comprising providing, by the data processing hardware, a transcription of the utterance for output.

3. The method of claim 1, wherein:

the accent library is selected from a plurality of accent libraries each associated with a different accent present within a particular language; and

a speaker of the utterance speaks the particular language as a second language.

4. The method of claim 1, further comprising:

obtaining, by the data processing hardware, device personalization data based on user interaction with the computing device,

wherein identifying a native language of a speaker of the utterance is further based on the device personalization data.

5. The method of claim 4, wherein the device personalization data comprises contextual application information.

6. The method of claim 1, further comprising:

determining, by the data processing hardware, demographic data of a speaker of the utterance,

wherein identifying the native language of the speaker of the utterance is further based on demographic data of the speaker of the utterance.

7. The method of claim 6, wherein the demographic data of the speaker includes a geographic location at which the speaker is located.

8. The method of claim 6, wherein the demographic data of the speaker is based on a country of an address stored in an address book of a computing device receiving the utterance.

9. The method of claim 1, further comprising:

determining, by the data processing hardware, a level of accuracy of a transcription of the utterance; and

selecting one or more additional accent libraries based on a level of accuracy of a transcription of the utterance,

wherein the speech recognition engine is further modified by the selected one or more additional accent libraries when generating the transcription of the utterance.

10. The method of claim 9, wherein as the level of accuracy of the transcription increases, the number of additional accent libraries selected increases.

11. A system, comprising:

data processing hardware; and

memory hardware in communication with the data processing hardware and having instructions stored thereon that, when executed on the data processing hardware, cause the data processing hardware to perform operations comprising:

receiving audio data of an utterance captured by a computing device;

identifying a native language of a speaker of the utterance based on a pitch or intonation within audio data of the utterance;

selecting an accent library comprising phonemes for pronunciations of words for a particular language based on a native language of a speaker of the recognized utterance; and

generating a transcription of the utterance using a speech recognition engine modified by the selected accent library.

12. The system of claim 11, wherein the operations further comprise providing a transcription of the utterance for output.

13. The system of claim 11, wherein:

the accent library is selected from a plurality of accent libraries each associated with a different accent present within a particular language; and

a speaker of the utterance speaks the particular language as a second language.

14. The system of claim 11, wherein the operations further comprise:

obtain device personalization data based on user interaction with the computing device,

wherein identifying a native language of a speaker of the utterance is further based on the device personalization data.

15. The system of claim 14, wherein the device personalization data comprises contextual application information.

16. The system of claim 11, wherein the operations further comprise:

determining demographic characteristic data of a speaker of the utterance,

wherein identifying the native language of the speaker of the utterance is further based on demographic data of the speaker of the utterance.

17. The system of claim 16, wherein the demographic data of the speaker includes a geographic location at which the speaker is located.

18. The system of claim 16, wherein the demographic data of the speaker is based on a country of an address stored in an address book of a computing device receiving the utterance.

19. The system of claim 11, wherein the operations further comprise:

determining a level of accuracy of a transcription of the utterance; and

selecting one or more additional accent libraries from a plurality of accent libraries based on a level of accuracy of a transcription of the utterance,

wherein the speech recognition engine is further modified by the selected one or more additional accent libraries when generating the transcription of the utterance.

20. The system of claim 19, wherein as the level of accuracy of the transcription increases, the number of additional accent libraries selected increases.

Background

Current speech recognition techniques are quite poor at recognizing speech when spoken with an accent. To address this problem, one partial solution tracks corrections made by a user in response to current techniques failing to correctly recognize words. This partial solution can be frustrating for users with accents because users often have to correct many incorrectly recognized words, often so many times that users completely forgo voice recognition, before these current techniques improve their recognition. Even for those users who are time consuming and frustrating, many current techniques still do not adequately recognize the user's speech when the user is accented.

Another partial solution for solving this problem requires the user to go to a dedicated user interface and speak a list of specific words. Requiring an accented user to find this special user interface and speak a list of words does not provide a superior user experience and thus would not be performed at all by the user. Further, requiring this effort from the user does not enable current techniques to identify accents sufficiently well. Still further, even if the user owning the device strives in this regard, it is unlikely to be performed by another user borrowing the owner's device, such as when the owner of the device is driving and the passenger is using the owner's device.

Drawings

Techniques and apparatus for recognizing accented speech are described with reference to the accompanying figures. Throughout the drawings, like numerals are used to reference like features and components:

FIG. 1 illustrates an example environment in which techniques for recognizing accented speech may be implemented.

FIG. 2 illustrates the example language and accent library of FIG. 1.

FIG. 3 illustrates an example method for recognizing accented speech using an accent library determined based on device data.

FIG. 4 illustrates an example method for modifying an accent library to more accurately recognize accented speech.

FIG. 5 illustrates an example method for recognizing speech at an application field based speech recognition level, which may use an accent library.

FIG. 6 illustrates an example application having an application field.

FIG. 7 illustrates various components of an example apparatus that may implement techniques for recognizing accented speech.

Detailed Description

Current techniques for recognizing accented speech tend to be quite poor in recognizing speech when accented. This disclosure describes techniques and apparatus for recognizing accented speech using an accent library, and in some embodiments, recognized words are arranged to be provided into an application field using different levels of speech recognition correction based on the application field.

The following discussion first describes an operating environment, followed by techniques that may be employed in such an environment, an example application having an application field, and proceeds to an example apparatus.

FIG. 1 illustrates an example environment 100 in which techniques for recognizing accented speech may be implemented. The example environment 100 includes a computing device 102, the computing device 102 having one or more processors 104, a computer-readable storage medium (storage medium) 106, a display 108, and an input mechanism 110.

Computing device 102 is shown as a smartphone with an integrated microphone 112 as one example of input mechanism 110. However, various types of computing devices and input mechanisms may be used, such as a personal computer with a separate independent microphone, a connection to a piconet with a microphone (e.g., Bluetooth)^TM) A cellular telephone with a headset, or a tablet and laptop computer with an integrated stereo microphone, to name a few.

The computer-readable storage media 106 includes an accent module 114, device data 116, mining data 118, and applications 120. The accent module 114 includes a language library 122 and one or more accent libraries 124. The accent module 114 may operate with, without, include, be integrated with, and/or supplement a speech recognition engine (not shown). The accent module 114 may be capable of recognizing accented speech, such as by determining, based on the device data 116, an accent library of accent libraries 124 that are combined with the language library 122 to recognize speech.

The language library 122 is associated with a language or dialect thereof, such as Australian English, United States (US) English, British (Royal) English, and the like. The language library 122 and known speech recognition engine may be operable to perform known speech recognition, although neither or both are required. Thus, in some embodiments, the accent module 114 uses one of the accent libraries 124 to supplement a known speech recognition engine that uses a known type of language library 122.

By way of example, consider FIG. 2, which illustrates the example language library 122 and accent library 124 of FIG. 1. Two example language libraries are shown here: australian english 204 and US english 206. Associated with each of these language libraries 204 and 206 are a number of accent libraries 208 and 210, respectively.

The accent library 208 includes eight examples (although the present technology contemplates more), including Australian (AU) english-national language 208-1, AU english-new south (N.S.) wils 208-2, AU english-New Zealand (NZ) orchids 208-3, AU english-NA crister chester 208-4, AU english-scuba-diver 208-5, AU english-inland 208-6, AU english-petas 208-7, and AU english-indonesia 208-8. As is clear from the name, each of these accent libraries is associated with a large language group (australian english) and accents present within that language group, whether it be a immigration recently speaking national language or a person participating in scuba diving.

Similarly, the accent library 210 includes eight examples: US English-national language 210-1, US English-Yue language 210-2, US English-Boston 210-3, US English-surfer 210-4, US English-hearing impairment 210-5, US English-rural 210-6, US English-southern 210-7, and US English-Alaska 210-8. Note that national spoken language libraries 208-1 and 210-1 may be different because each is associated with a different language library. However, whether English is spoken in the Australian dialect or the US dialect, there may be some common elements between accent libraries due to the common qualities of the speakers of the country. Note that these accent libraries are almost limitless in number and accents processed. Regional accents, small or large immigration groups, interests and sub-cultures, and even accents common to common physical characteristics, such as hearing impaired people have some commonality in accents.

In the example of fig. 2, each of the accent libraries 124 contains supplemental information and algorithms for use by the language library 122. Here the language library 122 is for a large group of languages (e.g., with more, mean or median, for a larger number of people), which is supplemented by one or more of the accent libraries 124. Although this example of FIG. 2 associates an accent library with a language library, the accent module 114 may forego the use of a language library or a known speech recognition engine. Instead, the accent module 114 may provide its own algorithms and engines without using other engines or libraries, instead relying on the accent library 124 rather than the language library 122, but including algorithms or information useful for recognizing the voices of a large number of people.

The accent module 114 may determine which of the accent libraries 124 to use for recognizing accented speech based on the device data 116 and/or the mining data 118 (both of fig. 1). The device data 116 may include device personal data 126 as well as data specific to the computing device 102. The data specific to the computing device 102 may include a date of manufacture or purchase of the computing device 102 (e.g., a recently released mobile phone or tablet) as well as information about the computing device 102 such as the manufacturer, hardware capabilities, etc.

Device personal data 126 includes data created or determined based on the user's interaction with computing device 102, such as the name of the contact, the installed application, the country or region of receipt of the message, the user's name, contact information, a non-standard keyboard (e.g., for a particular language other than the language for which computing device 102 is set), and contextual application information (e.g., search terms). Thus, the name of the contact may indicate the user's home country, or a non-standard type of keyboard may indicate that a language other than the language set for the computing device is the user's native language. Further, the receiving country or region of the message may include an address in a country where the language set for the computing device is not the most popular language, e.g., a receiving country in indonesia in the case of australian english being set (such as in the case of australian english 204 and AU english-indonesia 208-8 shown in fig. 2).

In more detail, an email or address in the user's contact address may indicate the user's nationality or ethnicity source (e.g., first or last name of the slavician). The address may indicate a birth location or current location of the user, as well as other details about the user that may be used to determine the accent library 124 for the user. The names in the email address lines or the body in those emails may indicate the nationality, the presence, the sub-culture, or the business of the user, or the interests of the user, of the user's friends. As further noted below, these interests may indicate accents, such as user interest in surfing, scuba diving, or cooking. Some words and how to say them may depend on these interests and thus be sub-cultural.

For example, a person participating in scuba diving may use the terms "ventilator" and "barotrauma," which may not be correctly identified if there is no accent library associated with scuba diving. Similarly, a person participating in surfing may use the terms "high flying foot", "back-rushing peak", or "a wave is stepped on all at once (closed out)", which may also not be correctly recognized from the user's speech. Finally, for cooking enthusiasts, "La createt," "barbeque," and "stew" may not be correctly identified without current technology.

The device personal data 126 may also include other information useful in determining accents and thus accent libraries, such as books in the user's electronic library in the slav language, news articles in the slav language, articles and books about the polish, weather channels saved in the polish, information about fishing in estonia, web search entries for accordion music, bill card music in the user's music library, and so forth.

The mining data 118 may also or instead be used by the accent module 114 to determine which of the accent libraries 124 to use for recognizing speech. Mining data 118 includes mining personal data 128, which may include any personal data that may be found about a user of computing device 102 over the Internet or otherwise. Thus, the mined personal data 128 may include the user's search terms, purchases, locations, demographics, revenue, and the like.

As noted, the computer-readable storage medium 106 also includes an application 120, such as an email application 130, a social networking application 132, or a spreadsheet application 134 of all of fig. 1. Each of the applications 120 includes one or more application fields 136, which are used in some embodiments to determine a level of speech recognition correction. By way of example, consider a spreadsheet application 134. Here only the numeric cell 138 and the general text cell 140 are examples of the application field 136. Only the numeric cells 138 may require more accurate text than the general text cells 140 and thus a different level of speech recognition correction.

FIG. 3 illustrates an example method 300 for recognizing accented speech using an accent library determined based on device data. The order in which the blocks of these and other methods are described is not intended to be construed as a limitation, and any number or combination of the blocks described herein in these and other methods may be combined in any order to implement the method, or an alternate method.

At block 302, device data for a computing device is received. The device data may be received in response to the active retrieval performed at block 302. Thus, using the environment 100 of fig. 1 as an example, the accent module 114 may retrieve the device data 116 at block 302, such as by searching for contact data on the computing device 102 and technical details about the computing device 102.

As noted in the above section, the device data 116 may include device personal data 126 as well as other non-personal data associated with the computing device 102. By way of an ongoing example, assume that device data 116 indicates that computing device 102 is a smartphone with significant computing power that was released only 30 days ago. This may be used, in part, to determine an appropriate accent library 124 based on demographics indicating that the user of this smartphone is an early adopter, proficient technique, and between 18 and 32 years of age, at least when recently released.

Assume that device personal data 126 includes contact names and addresses, indicating a statistically relevant number of asian family names and asian first names. This statistical relevance may be determined in various ways, such as by comparison to a contact list of typical people using the same language library 122. Thus, while the average number of Asian first names for a contact list of a user of the United States (US) English language library may be 1.3% and the Asian last names may be 11%, it is assumed herein that this user's contact list has 14% Asian first names and 29% Asian last names. The statistical analysis takes this statistical correlation into account based on whether it is one or more standard deviations from the mean. This indicates the likelihood that the user may not be native to english, or that the family members of the user are likely not to be a statistically relevant number of native to english, particularly asian first names, since asian first names are more likely to indicate first generation immigration than asian last names.

In addition to this information from the user's contact list, assume that the device personal data 126 indicates that the user's name is "Molly Chin," a large number of times and duration of trips to the beach, the purchase of surfing gear, and the user's residence in southern California.

At block 304, based on the received device data, an accent library is determined. This spoken library is determined for use in speech recognition. Continuing the ongoing embodiment, assume that the accent module 114 associates the device data 116 with a known accent associated with this type of device data, thereby determining that two different accent libraries 124 are possible, both US English-national language 210-1 and US English-surfer 210-4 of FIG. 2. Assume that the surfer accent library is determined to be more likely based on a younger age inferred for the user (e.g., early adopters, etc.), travel to the beach, name of the english person (Molly), based on the surfer's purchase, etc. In this ongoing example, the accent module 114 determines the accent library 124 based on the device data 116, although the accent module 114 may also or instead base this determination on the mined data 118 and information about speech previously received by the computing device 102.

At block 306, speech is received at the computing device. The speech may be received in various ways, such as the input mechanism 110 described above. Continuing the example in progress, assume that the user speaks the following to enter the text message "Jean, is it closed out? "

At block 308, speech is recognized based on the accent library. Concluding the ongoing example, in conjunction with the speech recognition engine, the accent module 114 uses the language library US English 206 and the accent library US English-surfer 210-4 selected based on the device data, as noted above. It is assumed here that without a spoken library, the speech recognition engine will "just, is it closed out" the speech of Molly? "identified as" Jean, is it close now? "however, because of the accent library US english-surfer 210-4, the accent module 114 acts to correctly recognize the voice of Molly as" just, is it close now? The accent module 114 then passes this text to the text field.

In this example, this recognition is due to the ability of the accent module 114 to select between multiple options for how to recognize Molly's speech, including an option that would be considered a low probability option for the current speech recognition engine if there were no accent library, relative to the other possible options of "close now", "hosed out", and "close". Here the accent library US english-the surfer 210-4 adds words, changes the probabilities of words and phrases, and modifies the algorithms to change how certain sounds are interpreted (e.g., the surfer has a different speech pattern, which is part of the accent, not just the words used).

Alternatively or additionally, the method 300 proceeds to block 310 and/or block 312-318. At block 310, the accent library is updated based on the corrected errors made during recognition of the speech. Block 310 may work in conjunction with or separately from method 400 as described below. In the example method 300 above, the correction recognized Molly's speech. If not, corrections by the user (Molly Chin) can be recorded and used to update the accent library.

At block 312, other speech is received at the computing device, the other speech received being from a different speaker than the speech received at 302. By way of example, assume that Molly handed her smartphone to her father because she is driving. Suppose that Molly asks her father for a good thailand restaurant. Also assume that her father is the native speaker in the national language and that English is his second language. Further, assume that like many native speakers, Molly's father uses intonation to distinguish words, while english speakers use intonation (a pitch pattern in a sentence). Further, suppose that Molly's father, like many speaking national speakers, has the problem of "I" sounds at the end of syllables. Thus, Molly's father pronounces "why" as "wiw," fly "as" flash, "and" pie "as" piw. Thus, when Molly's father wants the smartphone to Find Thai restaurants by speaking "Find Thai Restaurant," but, due to his accent, for a native US English speaker (or using only the speech recognition engine of the US English library), it sounds like "Find Tew Restaurant".

At block 314, another speech is dynamically determined to not be associated with the accent library determined at block 304. The accent module 114 determines in real time that the speaker is not Molly and thus the accent library US English-surfer 210-4 is not applicable when receiving the speech "Find Tew Restaurant". The accent module 114 may determine this based on "Tew" or other indicators, such as a change in pitch within the word "Restaurant," which is common to both national and cantonese speakers, or simply a speech history indication received from Molly that is not Molly. This can be performed in a number of ways, such as Molly having a generally high pitch and Molly's father not having this high pitch, a speaking speed difference between Molly and Molly's father, and so on.

At block 316, another accent library, or no accent library, is determined for the other speech. Continuing with this example, assume that accent module 114 determines that Molly's father is native to either national or Guangdong based on a change in pitch within the word "Restaurant". Further, assume that the accent module 114 determines that Molly's personal data indicates that she has friends and addresses that are more closely associated with a region of china (e.g., beijing) where chinese is the dominant language, rather than a region associated with cantonese (e.g., hong kong). As noted above, this information may have been determined at block 304.

At block 318, another speech is recognized with or without another accent library (as determined above). Ending the ongoing example, rather than incorrectly identifying this speech as "Find Two resurvants," the accent module 114 recognizes the parent's speech of Molly "Find Tew resurvants" as "Find Thai resurvants" by using the accent library US English-national language 210-1 of FIG. 2.

FIG. 4 illustrates an example method 400 for modifying an accent library to more accurately recognize accented speech.

At block 402, corrections to speech elements are received. This correction corrects for speech elements that were not correctly recognized using the accent library. The correction may be received from a remote computing device, although this is not required. As indicated in block 310, speech recognition using the accent library may be incorrect and then corrected by the user. One or more corrections associated with the accent library may be received, such as from thousands of remote computing devices (e.g., smartphones, laptops, tablets, desktop computers, etc.). The computing device may be computing device 102 of fig. 1, but in this embodiment is a server computer remote from computing device 102, and where the corrections are recorded and accent library 124 updated to improve recognition.

At block 404, the accent library is altered to provide an updated accent library that is capable of more accurately recognizing speech elements. To illustrate using one of The above examples, assume that The accent library US English-national language 210-1 incorrectly recognizes Molly's father's speech as "Find The Restaurant" instead of "Find Thai Restaurant". Assume also that Molly's father corrects the incorrect recognition to "Thai". This correction, as well as many other corrections of the same accent library as it is, may be sent to and received by the updating entity. The updating entity may be the accent module 114 on the computing device 102, or another accent module or other entity on the server computer.

At block 406, the updated accent library is provided to the one or more remote computing devices effective to enable the one or more remote computing devices to more accurately recognize the speech elements. Thus, using The updated accent library, The phonetic element "Tew" will more likely be correctly recognized as "Thai" rather than "The".

Further, device data may also be received from one or more remote computing devices, the device data associated with a user of the remote computing device, and a accent library is determined for speech recognition of speech from the user based on the device data. Thus, information about Molly for correcting the accent library US english-surfer 210-4 or information about Molly's father for correcting the accent library US english-national language 210-1 may be provided.

Updates to the appropriate accent library may then be customized for certain device data or other data. Indeed, over time this may act to provide sub-categories of accent libraries. Thus, a speaker, such as a person with similarities to Molly Chin, may receive updates from the US English-surfer 210-4 based on her similarities in age (18-30) and region (southern Calif.), while another speaker using the US English-surfer 210-4 will not receive, such as a male (age 45-60) living in a different region (Miami, Florida). By doing so, updates may be provided to the user based on whether the user or their computing device has one or more of the same devices or elements of mined data as the device or mined data of the remote computing device from which the corrections were received.

FIG. 5 illustrates an example method 500 for recognizing speech at an application field-based speech recognition level, which may use an accent library.

At block 502, speech is received at a computing device. This may be as set forth in the various examples above.

At block 504, a speech recognition correction level is determined based on the application field to which the recognized text is set to be provided. An example of this may be the example application field 136 of FIG. 1, namely the numeric-only cell 138 and the general text cell 140 of the spreadsheet application 134. As noted above, the accent module 114 may determine a level of speech recognition correction based on the application field, such as it may require highly accurate speech recognition or less accurate and/or faster recognition.

By way of example, consider fig. 6, which illustrates a user interface 602 of an example email application having application fields 604 and 606. Application field 604 is an address field and application field 606 is a body field. For example, assume that Molly Chin from the above example speaks "Surf Girl Seven Seven At Gee Mail Dot Com".

When a new email is opened to be sent to a friend, assume that the email application will first receive the recognized text into the email address field shown at application field 604. When speaking, and after the email address is complete, assume that the email application will receive the recognized text into the body of the email at application field 606. In this example, the accent module 114 determines that the maximum correction level should be used for the address field. In this case, the accent module 114 uses an appropriate accent library 124 or makes other refinements that improve accuracy. However, improving accuracy may come at the expense of time and computational resources (processor and battery) aspects of recognizing text, to name just a few. Thus, higher levels of speech correction may not always be appropriate.

For example, it is also noted that the accent module 114 may apply different levels of correction by determining to use zero, one, or multiple accent libraries 114 (such as both the national language and the surfer accent libraries). Further, accent module 114 may determine a level of correction that does not use or lacks use of accent library 124. For example, the accent module 114 may use a different language library 122 for some application fields, or an accent library 124 that points to spoken numbers rather than accents in normal speech. Thus, one of the language libraries 122 may point to speech that recognizes it as a number or for an address, and the other language library 122 points to speech that recognizes it as a conversation. In these and other ways set forth herein, the techniques may act to improve speech recognition.

At block 506, the received speech is recognized at a speech recognition correction level to produce recognized text. Thus, for the application field 604 (e-mail address field), the accent module 114 recognizes speech at a determined speech recognition level, here at a maximum level, using one or more accent libraries 124 and/or alternative language libraries 122 that point to the expected speech.

At block 508, the recognized words and other text are provided to the application field. Ending the ongoing example for Molly Chin, At block 508, instead of recognizing the speech "Surf Girl Seven Seven At Gee Mail Dot Com" as a word, the accent module 114 recognizes as a combination of words and text based on the accent library 124 and/or the language library 122, and also because it is the address field of the email, "At" is recognized as the "@" symbol. Thus, the speech is recognized as "surfgirl [email protected] gmail.

Although not required, in some embodiments, the techniques use a lower than maximum level of speech correction when the application field is the body of an email, blog, social networking portal, or word processing document. Conversely, for address fields, number-only fields in spreadsheets, telephone numbers, etc., the techniques may use a maximum level of speech correction and/or an alternative language library 122 or accent library 124.

Fig. 7 illustrates various components of an example device 700 including an accent module 114, the accent module 114 including or having access to other modules, the components being implemented in hardware, firmware, and/or software, and as described with reference to any of the previous fig. 1-6.

The example device 700 may be implemented as a fixed or mobile device that is one or a combination of the following: a media device, a computing device (e.g., computing device 102 of fig. 1), a television set-top box, a video processing and/or rendering device, an appliance device (e.g., an enclosed and sealed computing resource such as some digital video recorder or global positioning satellite device), a gaming device, an electronic device, a vehicle, and/or a workstation.

The example device 700 may be integrated with electronic circuitry, microprocessors, memory, input output (I/O) logic controls, communication interfaces and components, other hardware, firmware, and/or software necessary to operate the entire device. The example device 700 may also include an integrated data bus (not shown) that couples various components of the computing device for data communication between the components.

The example device 700 includes various components, such as an input-output (I/O) logic control 702 (e.g., for including electronic circuitry) and a microprocessor(s) 704 (e.g., a microcontroller or digital signal processor). The example device 700 also includes memory 706, which may be any type of Random Access Memory (RAM), low latency nonvolatile memory (e.g., flash memory), Read Only Memory (ROM), and/or other suitable electronic data storage. The memory 706 includes or has access to the accent module 114, the language library 122, and the accent library 124, and, in some embodiments, a speech recognition engine (not shown).

The example device 700 may also include various firmware and/or software, such as an operating system 708, which, among other components, may be computer-executable instructions held by the memory 706 and executed by the microprocessor 704. Example device 700 may also include other various communication interfaces and components, wireless lan (wlan) or wireless pan (wpan) components, other hardware, firmware, and/or software.

Other example capabilities and functions of these modules are described with reference to the elements shown in fig. 1 and 2. These modules, independently or in combination with other modules or entities, may be implemented as computer-executable instructions held by memory 706 and executed by microprocessor 704 to implement the various embodiments and/or features described herein. Alternatively or in addition, any or all of these components may be implemented as hardware, firmware, fixed logic circuitry, or any combination thereof, implemented in connection with I/O logic control 702 and/or other signal processing and control circuitry of example device 700. Further, some of these components may act separately from the device 700, such as when a remote (e.g., cloud-based) library executes the services of the accent module 114.

Although the invention has been described in language specific to structural features and/or methodological acts, it is to be understood that the invention defined in the appended claims is not necessarily limited to the specific features or acts described. Rather, the specific features and acts are disclosed as example forms of implementing the claimed invention.